diff --git a/src/system/kernel/scheduler/low_latency.cpp b/src/system/kernel/scheduler/low_latency.cpp
index a3eb28be4a..3e4365768b 100644
--- a/src/system/kernel/scheduler/low_latency.cpp
+++ b/src/system/kernel/scheduler/low_latency.cpp
@@ -30,9 +30,6 @@ has_cache_expired(Thread* thread)
 {
 	ASSERT(!gSingleCore);
 
-	if (thread_is_idle_thread(thread))
-		return false;
-
 	scheduler_thread_data* schedulerThreadData = thread->scheduler_data;
 	ASSERT(schedulerThreadData->previous_core >= 0);
 
diff --git a/src/system/kernel/scheduler/power_saving.cpp b/src/system/kernel/scheduler/power_saving.cpp
index 75d754f178..aa960e49cd 100644
--- a/src/system/kernel/scheduler/power_saving.cpp
+++ b/src/system/kernel/scheduler/power_saving.cpp
@@ -36,9 +36,6 @@ has_cache_expired(Thread* thread)
 {
 	ASSERT(!gSingleCore);
 
-	if (thread_is_idle_thread(thread))
-		return false;
-
 	scheduler_thread_data* schedulerThreadData = thread->scheduler_data;
 	ASSERT(schedulerThreadData->previous_core >= 0);
 
@@ -46,19 +43,6 @@ has_cache_expired(Thread* thread)
 }
 
 
-static bool
-try_small_task_packing(Thread* thread)
-{
-	ReadSpinLocker locker(gCoreHeapsLock);
-
-	int32 core = sSmallTaskCore;
-	return (core == -1 && gCoreLoadHeap->PeekMaximum() != NULL)
-		|| (core != -1
-			&& get_core_load(&gCoreEntries[core]) + thread->scheduler_data->load
-				< kHighLoad);
-}
-
-
 static int32
 choose_small_task_core(void)
 {
@@ -107,24 +91,27 @@ choose_core(Thread* thread)
 {
 	CoreEntry* entry;
 
-	if (try_small_task_packing(thread)) {
-		// try to pack all threads on one core
-		entry = &gCoreEntries[choose_small_task_core()];
+	int32 core = -1;
+	// try to pack all threads on one core
+	core = choose_small_task_core();
+
+	if (core != -1
+		&& get_core_load(&gCoreEntries[core]) + thread->scheduler_data->load
+			< kHighLoad) {
+		entry = &gCoreEntries[core];
 	} else {
 		ReadSpinLocker coreLocker(gCoreHeapsLock);
-		if (gCoreLoadHeap->PeekMinimum() != NULL) {
-			// run immediately on already woken core
-			entry = gCoreLoadHeap->PeekMinimum();
-		} else {
+		// run immediately on already woken core
+		entry = gCoreLoadHeap->PeekMinimum();
+		if (entry == NULL) {
 			coreLocker.Unlock();
 
 			entry = choose_idle_core();
 
-			coreLocker.Lock();
-			if (entry == NULL)
-				entry = gCoreLoadHeap->PeekMinimum();
-			if (entry == NULL)
+			if (entry == NULL) {
+				coreLocker.Lock();
 				entry = gCoreHighLoadHeap->PeekMinimum();
+			}
 		}
 	}
 
@@ -138,9 +125,6 @@ should_rebalance(Thread* thread)
 {
 	ASSERT(!gSingleCore);
 
-	if (thread_is_idle_thread(thread))
-		return false;
-
 	scheduler_thread_data* schedulerThreadData = thread->scheduler_data;
 	ASSERT(schedulerThreadData->previous_core >= 0);
 
@@ -151,13 +135,16 @@ should_rebalance(Thread* thread)
 	if (coreLoad > kHighLoad) {
 		ReadSpinLocker coreLocker(gCoreHeapsLock);
 		if (sSmallTaskCore == core) {
-			if (coreLoad - schedulerThreadData->load < kHighLoad)
-				return true;
-
+			sSmallTaskCore = -1;
 			choose_small_task_core();
+			if (schedulerThreadData->load > coreLoad / 3)
+				return false;
 			return coreLoad > kVeryHighLoad;
 		}
 
+		if (schedulerThreadData->load >= coreLoad / 2)
+			return false;
+
 		CoreEntry* other = gCoreLoadHeap->PeekMaximum();
 		if (other == NULL)
 			other = gCoreHighLoadHeap->PeekMinimum();
@@ -165,10 +152,15 @@ should_rebalance(Thread* thread)
 		return coreLoad - get_core_load(other) >= kLoadDifference / 2;
 	}
 
+	if (coreLoad >= kMediumLoad)
+		return false;
+
 	int32 smallTaskCore = choose_small_task_core();
 	if (smallTaskCore == -1)
 		return false;
-	return smallTaskCore != core;
+	return smallTaskCore != core
+		&& get_core_load(&gCoreEntries[smallTaskCore])
+				+ thread->scheduler_data->load < kHighLoad;
 }
 
 
diff --git a/src/system/kernel/scheduler/scheduler.cpp b/src/system/kernel/scheduler/scheduler.cpp
index b1bcef7d15..8fb6cf970f 100644
--- a/src/system/kernel/scheduler/scheduler.cpp
+++ b/src/system/kernel/scheduler/scheduler.cpp
@@ -120,7 +120,8 @@ CoreEntry::CoreEntry()
 	fCPUCount(0),
 	fThreadCount(0),
 	fActiveTime(0),
-	fLoad(0)
+	fLoad(0),
+	fHighLoad(false)
 {
 	B_INITIALIZE_SPINLOCK(&fCPULock);
 	B_INITIALIZE_SPINLOCK(&fQueueLock);
@@ -264,6 +265,7 @@ dump_cpu_heap(int argc, char** argv)
 {
 	kprintf("core load\n");
 	dump_core_load_heap(gCoreLoadHeap);
+	kprintf("\n");
 	dump_core_load_heap(gCoreHighLoadHeap);
 
 	for (int32 i = 0; i < gCoreCount; i++) {
@@ -376,23 +378,32 @@ update_load_heaps(int32 core)
 		return;
 
 	if (newKey > kHighLoad) {
-		if (oldKey <= kHighLoad) {
+		if (!entry->fHighLoad) {
 			gCoreLoadHeap->ModifyKey(entry, -1);
 			ASSERT(gCoreLoadHeap->PeekMinimum() == entry);
 			gCoreLoadHeap->RemoveMinimum();
 
 			gCoreHighLoadHeap->Insert(entry, newKey);
+
+			entry->fHighLoad = true;
 		} else
 			gCoreHighLoadHeap->ModifyKey(entry, newKey);
-	} else {
-		if (oldKey > kHighLoad) {
+	} else if (newKey < kMediumLoad) {
+		if (entry->fHighLoad) {
 			gCoreHighLoadHeap->ModifyKey(entry, -1);
 			ASSERT(gCoreHighLoadHeap->PeekMinimum() == entry);
 			gCoreHighLoadHeap->RemoveMinimum();
 
 			gCoreLoadHeap->Insert(entry, newKey);
+
+			entry->fHighLoad = false;
 		} else
 			gCoreLoadHeap->ModifyKey(entry, newKey);
+	} else {
+		if (entry->fHighLoad)
+			gCoreHighLoadHeap->ModifyKey(entry, newKey);
+		else
+			gCoreLoadHeap->ModifyKey(entry, newKey);
 	}
 }
 
diff --git a/src/system/kernel/scheduler/scheduler_common.h b/src/system/kernel/scheduler/scheduler_common.h
index f379ba9e36..e92b3a6763 100644
--- a/src/system/kernel/scheduler/scheduler_common.h
+++ b/src/system/kernel/scheduler/scheduler_common.h
@@ -36,6 +36,7 @@ const bigtime_t kCacheExpire = 100000;
 const int kLowLoad = kMaxLoad * 20 / 100;
 const int kTargetLoad = kMaxLoad * 55 / 100;
 const int kHighLoad = kMaxLoad * 70 / 100;
+const int kMediumLoad = (kHighLoad + kTargetLoad) / 2;
 const int kVeryHighLoad = (kMaxLoad + kHighLoad) / 2;
 
 const int kLoadDifference = kMaxLoad * 20 / 100;
@@ -83,6 +84,7 @@ struct CoreEntry : public MinMaxHeapLinkImpl<CoreEntry, int32>,
 	bigtime_t	fActiveTime;
 
 	int32		fLoad;
+	bool		fHighLoad;
 } CACHE_LINE_ALIGN;
 typedef MinMaxHeap<CoreEntry, int32> CoreLoadHeap;